NammaSatya

Inspiration

Bengaluru residents get a constant stream of civic claims on WhatsApp and Telegram, including metro shutdowns, water cuts, BBMP notices, plastic bans, road closures, and more. By the time these messages reach people, the original fact is often exaggerated, distorted, or completely invented.

We wanted to build a fast, trustworthy way for people to check these claims before forwarding them. Generic chatbots can hallucinate local civic facts, and official websites are hard to search under pressure, so we built NammaSatya as a grounded, citation-backed truth-checking agent for Bengaluru.

What it does

NammaSatya lets users paste a viral civic claim and instantly returns a verdict with evidence.

The verdict can be:

• SUPPORTED: the claim matches verified sources
• REFUTED: verified sources directly contradict the claim
• UNVERIFIED: no indexed source covers the claim
• MANGLED: the claim has a true core, but was distorted while being forwarded

Every answer includes a confidence score, a short explanation, direct citations, source URLs, excerpts, dates, and freshness information.

For example, if someone claims that the entire Purple Line is shutting down, NammaSatya can identify whether the real source only says a specific section is suspended for a limited time, and mark the viral claim as MANGLED instead of simply true or false.

How we built it

We built a retrieval-grounded fact-checking pipeline:

Raw claim
→ Sanitizer
→ Search query extraction
→ Elasticsearch hybrid search
→ LLM reranking
→ Claude verdict generation
→ Structured verdict with citations

Built With

• aws-bedrock
• claude-haiku
• claude-sonnet
• elastic-agent-builder
• elastic-open-crawler
• elasticsearch
• elser
• kibana
• mcp
• python